Shadow emoval approaches improvement under challengeing conditions

Abstract

Shadow removal is with no doubt an interesting topic, the benefits of developing a way to remove shadows from images are intriguing. However, removing shadows from single and multiple images is a challenging problem and producing high-quality images that the naked eye can't distinguish from a real shadowless scene is on another level of difficulty. Shadows present a complex phenomenon, especially in digital images, where their complexity can be affected by several factors in the image acquisition, the storage, the processing and even in the post-processing phase which can introduce noise and artifacts rendering shadow removal harder. Over the years, numerous approaches to shadow removal were proposed for single and multiple images. However, many methods fail to handle some aspects of the shadows' complexities in shadow removal, these can significantly affect the performance and the final results, and that can be more noticeable when removing non-uniform or irregular soft shadows. In this work we focused on three major points: first, we provided the different methods used in literature as well as an enhanced survey of the problems and challenges that affect the process of shadow removal. Second, two state-of-the-art approaches for removing shadows from single color images were presented. The mathematical background was reviewed to show how the proposed techniques are robust in removing shadows cast on different surfaces. Third, we proposed two algorithms for shadow detection and generation, which can effectively detect and generate artificial shadow respectively. The experimental results were given in order to demonstrate the capabilities and robustness of the proposed approaches in new and more challenging lights; their limitations were discussed to highlight the different areas for future work.